Next-generation motion analysis and motion feedback for sports using omnidirectional cameras

Objectives: Novel three-dimensional (3D) motion analysis methods using omnidirectional cameras are developed and a new framework is presented to integrate this analysis with immersive Virtual Reality (VR) motion feedback. This study addresses the limitations of sports motion analysis that relies on narrow-field-of-view conventional cameras. Design: Employing a novel methodological and experimental design, new mathematical operations tailored to the spherical projection of omnidirectional cameras were developed and validated. Subsequently, the dual application of the design was demonstrated in quantitative analysis and experiential feedback. Methods: New 3D reconstruction algorithms, inspired by the Direct Linear Transformation (DLT) and Non-Linear Transformation (NLT) principles but adapted for omnidirectional imagery, were developed. These methods utilize the longitude and latitude of celestial sphere images to determine the camera ray direction, thereby enabling the determination of the camera position/orientation (DLT-inspired method) or the relative position/orientation of the cameras (NLT-inspired method). A separate purely geometric method for minimal-calibration 2D position determination using two omnidirectional cameras was also proposed. Omnidirectional imagery data were then used to generate immersive 360° VR content for motion feedback using a Head-Mounted Display (HMD). Results: The DLT-inspired 3D reconstruction method achieved an error of 0.22% relative to the calibrated space volume, and the NLT-inspired method achieved an error of 0.34%, both comparable to the accuracy of the gold standard methods. These methods drastically reduce the required number of cameras and technical complexity. Furthermore, omnidirectional recordings were successfully transformed into immersive VR content, enabling an embodied re-experience of movement. Conclusions: Omnidirectional cameras have successfully overcome the field-of-view and complexity limitations of traditional DLT/NLT methods, achieving high-accuracy 3D motion analysis with minimal camera units. The integration of this precise analysis with immersive VR motion feedback results in a powerful, unified framework that accelerates motor-skill acquisition and performance enhancement. Thus, this study paves the way for the next generation of real-time mobile motion intelligence systems in sports science.